4
Scientific Considerations

Implicit in the guiding principles developed in Chapter 3 is the assumption that there are meaningful differences between the sexes, and that—at least in some instances—the results of male-only studies cannot be reliably or safely generalized to women. This chapter examines the scientific evidence for that assumption and its implications for the design of clinical studies. Both of these issues have consequences for a policy of inclusion, as mandated by the NIH Revitalization Act of 1993.

There is a general belief among clinical researchers that, in most situations, women and men will not differ significantly in their response to treatment. The evidence to support this belief is not easily assembled, however, and there are countervailing concerns that gender differences have been insufficiently studied. Some of the known gender differences in response to treatments are related to physiological differences between the genders. Important examples include hormonal differences, particularly the variation in drug response by women during different stages of the menstrual cycle, and pharmacokinetic effects such as differential rates of drug absorption and excretion. Other differences are psychosocial in origin or are mediated by tendencies of men and women to act differently with respect to health care. In cases where there is substantial evidence of a qualitative or large quantitative difference in response by gender, the weight of evidence supports a policy of including both genders in sufficient number to permit subgroup analyses, except in studies involving conditions or treatments that affect only one gender.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 84
--> 4 Scientific Considerations Implicit in the guiding principles developed in Chapter 3 is the assumption that there are meaningful differences between the sexes, and that—at least in some instances—the results of male-only studies cannot be reliably or safely generalized to women. This chapter examines the scientific evidence for that assumption and its implications for the design of clinical studies. Both of these issues have consequences for a policy of inclusion, as mandated by the NIH Revitalization Act of 1993. There is a general belief among clinical researchers that, in most situations, women and men will not differ significantly in their response to treatment. The evidence to support this belief is not easily assembled, however, and there are countervailing concerns that gender differences have been insufficiently studied. Some of the known gender differences in response to treatments are related to physiological differences between the genders. Important examples include hormonal differences, particularly the variation in drug response by women during different stages of the menstrual cycle, and pharmacokinetic effects such as differential rates of drug absorption and excretion. Other differences are psychosocial in origin or are mediated by tendencies of men and women to act differently with respect to health care. In cases where there is substantial evidence of a qualitative or large quantitative difference in response by gender, the weight of evidence supports a policy of including both genders in sufficient number to permit subgroup analyses, except in studies involving conditions or treatments that affect only one gender.

OCR for page 84
--> These true gender differences (and differences associated with gender, e.g., weight) have implications for the design of clinical trials, the subset of clinical studies that provides the most rigorous and reliable test of the effectiveness and safety of new drugs and treatments. For example, greater heterogeneity among research subjects may permit the investigator to spot trends that might otherwise be missed, even if the numbers are too small for statistically reliable subgroup analysis. At the same time, greater homogeneity among research subjects reduces unexplained variance. EVIDENCE OF GENDER DIFFERENCES This report, originating as it does from concerns about insufficient attention being directed toward identifying and understanding gender differences, of necessity highlights diseases and treatments that can differ by gender in a variety of ways. Differences can arise from a range of factors, both biological (e.g., the effect of endogenous or exogenous hormones or gender-related differences in body mass, etc.) and psychosocial (e.g., gender-related differences in behaviors such as smoking or substance abuse). The question that must be addressed is to what extent are gender differences per se clinically meaningful in the treatment of conditions involving both genders? Most clinical researchers and clinicians would argue that women and men do not respond significantly differently to the presence of disease or the effect of treatment. Even for diseases where women and men differ significantly in the likely time of onset (such as heart disease), they will usually respond in much the same way to treatment and experience a similar evolution of the disease. The underlying reasons for this belief are rooted in several observations regarding health problems relevant to both men and women: for the majority of drug treatments, efficacy and safety do not depend on such factors as body mass, adipose tissue, hormones, or other factors associated with gender. Treatments by surgical procedure for diseases associated with both genders seldom differ because the patient is a woman instead of a man; and to the extent that women may be treated differently, it is because of factors associated with gender but not specific to gender, such as bone mass or organ size. Finally, a long history of nonhuman research—ranging from work with bacteria to research with mammals—supports the conclusion that subgroup differences are rare. Most treatments and disease processes are thus thought to be insensitive to gender per se. Nevertheless, the evidentiary base for quantifying these claims in humans is weak because the relevant data have not been organized into an accessible format and the claim is seldom questioned. At the same time, concern is mounting among both scientific observers and lay representatives that researchers and clinicians may be too quick to

OCR for page 84
--> assume there are no differences between women and men, rather than test for gender-related effects. Much of the controversy surrounding the issue of women's participation in clinical studies has centered on differential responses to drugs between men and women. Significant gender differences in drug response have not been detected in the majority of cases, but where they are detected they can be important. Therefore, it becomes important for clinical investigators to ascertain under what conditions such gender differences are likely to occur and to design clinical studies accordingly. The following sections summarize some of the literature on documented gender differences, primarily with respect to differences in drug response. Body Size, Composition, and Metabolism Men and women differ in body size, body composition, and metabolism (Table 4-1). On average, women are smaller than men in weight, height, and surface area (Silvaggio and Mattison, 1993). This may affect drug dosing, which may be more appropriately based on body weight (or surface area) than on a fixed dose. It is of interest that most adult dosing is done on a fixed dose, not based on weight or surface area. For example, if a drug is administered on the basis of body weight (say 10 mg/kg), then a typical adult male will receive a dose of 700 mg and a typical adult female 570 mg. If the same drug is given on the basis of surface area (say, 380 mg/m), then the average adult male will receive a dose of 703 mg and the average female 608 mg. If weight or surface area are not taken into consideration, however, and men and women are given the same dose (700 mg, for example), then the average male will get a dose of 10.0 mg/kg and the average female 12.2 mg/kg. If the drug has minimal toxicity or a wide therapeutic index, these differences in dosage may be of little consequence. If the therapeutic index is narrow, however, or the toxicity severe, these differences may be of critical importance. Compared with men, women also have a lower ratio of lean body mass to adipose (fatty) tissue (Yonkers et al., 1992). This difference in body composition may affect drug disposition—for example, the water content and metabolism of adipose tissue differs from that in muscle tissue (Silvaggio and Mattison, 1993). Lipid-soluble drugs such as the benzodiazepines would be expected to have a greater volume of distribution in women (on the basis of body weight or surface area), which would affect the appropriate therapeutic dose. For all ages after sexual maturation, metabolism (as measured by basal metabolic rates) is higher in men. Drug metabolism differences by gender have been poorly studied. A few drugs, however, demonstrate gender differences in metabolism, including nicotine, acetylsalicylic acid (aspirin), and heparin (an anticoagulant) (Silvaggio and Mattison, 1993).

OCR for page 84
--> TABLE 4-1 Representative Values of Weight, Height, Surface Area, and Caloric Intake for Adult Males and Females Sex Weight (kg) Height (cm) Surface Area (m2) Caloric Intake Kilocalories/Kilograma Kilocalories/Meter Male 70.0 178 1.85 3,000 43 1,620 Female 57.0 163 1.60 2,300 40 1,440 a Kilocalories per kilogram are given to indicate the significant differences in metabolic rate when calculated on the basis of body weight. SOURCE: NRC, 1993.

OCR for page 84
--> TABLE 4-2 Concentrations of Steroids/Hormones in Males and Females Item by Gender Concentration Estrogen (urinary) Males 5-40 n mole/24 h Females <10 years 0.5 follicular 20-150 mid-cycle 60-300 luteal phase 45-290 postmenopausal 10-55 Progesterone (plasma) Males <5 n mole/L Females 15-77 n mole/L Prolactin (plasma) Males <450 U/L Females <600 U/L Testosterone Males 14-42 n mole/L Females 1-2.1   SOURCE: Wetherall et al., 1988. Another obvious difference between men and women is the presence of hormones such as estrogen, progesterone, prolactin, and testosterone (Table 4-2). These hormonal differences are important for establishing and maintaining a range of gender-dependent physiological characteristics. They may also modify the pharmacokinetics and pharmacodynamics of selected drugs. Gender differences may also be found in other commonly measured laboratory tests such as serum iron, uric acid, creatinine phosphokinase, and gamma glutamyl transpeptidase, all of which are important in distinguishing the normal from the abnormal in selected disease states (Table 4-3). TABLE 4-3 Concentrations (μ mole/liter) of Commonly Measured Tests in Males and Females Group Serum Iron Uric Acid Creatine Phosphokinase Gamma Glutamyl Transpeptidase Males 14-31 210-480 25-195 11-51 Females 11-30 150-390 25- 170 7-33   SOURCE: Wetherall et al., 1988

OCR for page 84
--> Gender and Aging Men and women over the age of 65 currently make up 13 percent of the total population in this country, a percentage that is expected to grow significantly in the coming decades (Olshansky et al., 1993). Because women live longer than men by an average of 7 years, they presently constitute a majority (59 percent) of those over 65 and nearly 75 percent of those over the age of 85. While many of the medically relevant differences that exist between younger men and women persist into older age, men and women over 65 also experience gender-associated health problems that are unique to their age group. Frequently associated with advanced age, for instance, is the increased use of medications. Men and women over the age of 65 consume 30 percent of all prescription drugs sold in the United States (NIA, 1992). Of this amount, women consume 20 percent more drugs than men (National Medical Expenditure Survey, 1987). For women over 65, this translates into an average consumption of 5.7 prescriptions a year, not including the average consumption of 3.2 over-the-counter drugs a year by this group (National Center for Health Statistics, 1985). One-quarter of women over 65 consume as many as 21 different medications in a year; one-fifth of men over 65 consume this many medications in a year (National Medical Expenditure Survey, 1987). Women over 65 suffer more adverse events related to medication than do men in the same age group, although these events are more likely to result in death for males. For both genders, cardiac, antihypertensive, and nonsteroidal anti-inflammatory agents are the most common causes of these events (Tanner et al., 1989). Adverse events may also occur in both sexes with the use of water-soluble drugs such as lithium. Although older men and women both experience decreases in lean body mass and increases in fat tissue as a fraction of body weight, these changes may be more pronounced in women, who tend to have more body fat than males in youth and middle age. As a result, drugs such as lithium may have a more immediate toxic effect in older women (Everitt and Avorn, 1986). In addition, lipophilic drugs such as phenothiazines and many benzodiazepines may have a more prolonged effect in older women. Men and women over 65 also differ with respect to the diseases and conditions that commonly affect them. For example, older women are affected by rheumatoid arthritis three times more frequently than men, while men over 65 are affected by gout in significantly greater numbers (NIA, 1992). Osteoporosis is far more common in women than men, affecting slightly over 73 percent of women aged 65-69 and 89 percent of women over age 75 (NIH, 1992). Women develop osteoporosis with advanced age as a result of decreased hormone levels, too little calcium in the diet, inac-

OCR for page 84
--> tivity, and perhaps heredity; men appear to be protected from the disease by denser bone structure and other factors (NIA, 1992). In addition to osteoporosis, women over 65 experience other chronic conditions with greater frequency than men in the same age group, including digestive disorders and thyroid diseases. Mental health problems also afflict older women to a greater extent than older men. Women over 65 experience depression four times as often as men; anxiety disorders, sleep disorders, mania, and late-onset psychosis also occur more frequently in older women (NIH, 1992). Neurodegenerative diseases such as Alzheimer's disease and certain movement disorders also affect women disproportionately, although it is unclear whether this reflects brain differences or simply differences in survival. Evidence for the latter hypothesis is strong: the majority of Alzheimer's disease patients are over the age of 80, an age when women outnumber men nearly two to one (NIH Advisory Committee on Women's Health Issues, 1988). Just as sources of morbidity differ between men and women over 65, so too do causes of mortality. Older men die from heart disease and malignant neoplasms more frequently than do older women, who die more often from cerebrovascular disease (National Center for Health Statistics, 1993). Nevertheless, these three diseases remain the top killers of all persons over the age of 65. Behavioral and Psychosocial Differences Men and women also differ in a number of psychosocial variables that can affect disease risk, treatment, or prevention. These variables mainly pertain to gender roles and lifestyle (Rodin and Ickovics, 1990). There are stresses associated with the multiple roles women typically assume in taking on the responsibilities of balancing work and family, but the evidence of negative effects from this source of stress on women's health is inconclusive (Horton, 1992). Women also are more likely to experience domestic violence leading to physical and psychological injuries (Rodin and Ickovics, 1990). Twice as many women as men experience major depression, and a number of psychosocial factors have been found to be risk factors for depression in women, including imbalances in perceived control over one's life, social support, and sense of accomplishment and independence (McGrath et al., 1990; Horton, 1992). Gender differences in lifestyle may also affect health. Women tend to get less regular exercise than men, and a sedentary lifestyle has been linked to cardiovascular disease in men (National Women's Health Resource Center, 1990). More men than women drink alcohol and smoke tobacco, but increased consumption of both substances by women over the past few decades poses serious risks to women's health. Because cigarette smoking

OCR for page 84
--> is an important risk factor for numerous diseases—including heart disease, lung cancer, and chronic obstructive pulmonary disease—the rise in smoking among adolescent and young women is of special concern. There is increasing evidence that even moderate use of alcohol increases the risk for breast cancer in women by 50 percent (Rodin and Ickovics, 1990). Cultural emphasis on thinness and beauty in women translate into a higher prevalence of eating disorders, such as anorexia nervosa and bulimia (Horton, 1992) and high use of over-the-counter diet pills (women constitute 90 percent of users) (Rodin and Ickovics, 1990). The poor nutrition in childhood and adolescence that results from such dieting creates a greater risk for osteoporosis in later life. The desire for thinness has also been implicated in women smokers' reluctance to quit, because the average quitter gains about five pounds (Horton, 1992). The pressure to be thin may also affect women's compliance with drug regimens when the side effects include weight gain (National Women's Health Resource Center, 1990). Endogenous Hormones Not all gender differences are merely a matter of degree; some are specific to one gender. From menarche to menopause, women undergo cyclical physiological changes, with a duration of approximately 28 days, known as the menstrual cycle. The menses are characterized by low levels of the hormones estrogen and progesterone. During the follicular phase, the level of estrogen rises and the endometrium thickens to enable a fertilized ovum to implant if conception occurs during that cycle. During the luteal phase, following ovulation, the level of estrogen declines and the level of progesterone rises. If fertilization and implantation do not occur, the level of progesterone declines and the menses begin a new cycle. Endogenous hormonal changes in menstruating women can affect drug disposition, but few studies have examined the impact of changing hormonal concentrations on drug metabolism across the different phases of the menstrual cycle (GAO, 1992; NRC, 1993). One recent report noted the importance of varying the dose of an antidepressant over the menstrual cycle to achieve optimal benefit and minimal side effects (Jensvold et al., 1992). Thus, although rarely studied, there is some evidence that while constant drug dosing may be appropriate for males, females may benefit from variable dosing tailored to their menstrual cycles. Endogenous hormones may also affect the success of some types of surgical treatments. For example, recent studies suggest a potential link between the timing (within the menstrual cycle) of surgery for breast cancer and the rate of recurrence and survival (NRC, 1993). Multivariate analyses, controlling for such variables as tumor size and number of lymph nodes

OCR for page 84
--> involved, suggest that surgery during the early luteal phase of the cycle enhances survival, although the exact reason is undetermined. Pregnancy and Lactation Numerous physiological changes occur during pregnancy, and some of these changes persist during lactation. Beyond changing the size, shape, and center of gravity in the body, changes occur in the pulmonary, cardiovascular, renal, gastrointestinal, and hepatic systems. These changes can alter the body's disposition of drugs, including absorption, distribution, metabolism, and elimination (Hytten and Leitch, 1971; Hytten and Chamberlain 1980; Mattison and Jelovsek, 1991; Mattison et al., 1992; Mattison 1986, 1990). Table 4-4 lists some of these physiological changes and their pharmacokinetic effects. Some of the changes that occur during pregnancy, such as increased plasma volume, body weight, and body fat, can decrease the concentration TABLE 4-4 Changes During Pregnancy that May Alter Pharmacokinetics Pharmacokinetic Parameter Change Pharmacokinetic Impact Absorption     Gastric emptying time Increased Increased absorption and/or metabolism Intestinal motility Decreased Increased absorption Pulmonary function Increased Increased absorption and/or elimination Cardiac output Increased Increased distribution rate Blood flow to the skin Increased Increased transdermal absorption Distribution     Plasma volume Increased Increased volume of distribution Total body water Increased Increased volume of distribution Plasma proteins Decreased Decreased volume of distribution, decreased binding capacity Body fat Increased Increased volume of distribution, increased reservoir for lipid-soluble xenobiotics Metabolism     Hepatic metabolism +/- +/- Metabolic alteration and elimination Extra-hepatic metabolism +/- +/- Metabolic alteration and elimination Plasma proteins Decreased Increased metabolic alteration and elimination Excretion     Renal blood flow Increased Increased renal elimination Glomerular filtration rate Increased Increased renal elimination Pulmonary function Increased Increased pulmonary elimination Plasma proteins Decreased +/- Elimination   SOURCE: Modified from Mattison, 1986.

OCR for page 84
--> of a drug in the body, thereby decreasing its therapeutic effect. Pregnant women with epilepsy, for example, typically require higher doses of phenytoin (by 50 percent or more) to avoid seizures than nonpregnant women. Other changes, such as decreased intestinal motility and decreased gastric emptying time, can cause a drug to be absorbed in greater amounts than when a woman is not pregnant. These circumstances may warrant decreasing dosage of some drugs during pregnancy. In addition to changing maternal drug disposition, both the placenta and fetus can contribute to modified drug disposition (Gillette, 1977; Pelkonen, 1980). These effects, however, are thought to be small compared with maternal effects on the drug. It may be ideal for women to avoid medication altogether while pregnant, but this can be difficult (see also Chapter 7). A wide array of maternal, fetal, and pregnancy-associated conditions can require treatment during pregnancy, and as many as 75 percent of pregnant women use prescription or over-the-counter drugs—hence the importance of understanding the pharmacokinetic changes that occur during pregnancy. Many of the physiological changes that occur with pregnancy persist for some time after childbirth, so their impact on a nursing infant must also be considered in drug dosing. Factors that determine the concentration of drugs in breast milk include dosing interval, frequency of breastfeeding, and lipid solubility. Because breast milk is rich in lipids (3 to 5 percent by volume), lipid-soluble drugs are preferentially found in breast milk. Although most drugs taken by the mother are found in breast milk, the dose to the infant is typically small. Exogenous Hormones Hormonal Contraceptives Data from a 1988 survey indicate that 31 percent of U.S. women between the ages of 15 and 44 use oral contraceptives (OCs) (Mosher, 1990). Most OCs are pills that contain synthetic hormones, usually estrogen and progestin, although some OCs contain only a progestin compound. In addition to being highly effective in preventing pregnancy, OCs also reduce the risk of ovarian and endometrial cancer. Two other long-acting hormonal contraceptives that contain only progestins have recently become available—Norplant® and DepoProvera®—and close to a million U.S. women are now using these methods. Despite their many health benefits, OCs also have adverse effects. For example, OCs have been found to increase the risk of coronary heart disease, particularly acute myocardial infarction, a risk that is compounded in smokers. Most of the studies linking OCs to heart disease, however, are based on women who took pills containing much higher doses of estrogen

OCR for page 84
--> and progestin than are contained in pills used today. More research is needed to assess the effects of lower-dose OCs on the risk of cardiovascular disease in smokers and nonsmokers (National Women's Health Resource Center, 1990). Oral contraceptives can also influence the pharmacokinetics and pharmacodynamics of other drugs. For example, the ethinyl estradiol component of OCs is responsible for reducing hepatic metabolism of many drugs, thereby increasing plasma concentrations (e.g., prednisolone, antipyrine, imipramine, diazepam, chlordiazepoxide, phenytoin, caffeine, and cyclosporine) (Teichmann, 1990; NRC, 1993). For other drugs, OCs can induce drug metabolism, thus lowering plasma concentrations (as with acetaminophen, morphine, lorazepam, and oxazepam) (NRC, 1993). Drug interactions can also reduce the efficacy of OCs themselves; anticonvulsants in particular are known to reduce OC concentrations and effectiveness. Menopause and Hormone Replacement Menopause, the time in a woman's life when the production of estrogen declines and menstruation ceases, usually occurs sometime between the midforties and early fifties. Many signs and symptoms accompany the onset of menopause, including hot flashes, irritability, and sleep disturbances (Bush, 1992). Estrogen replacement therapy has been recommended to reduce or eliminate these unpleasant symptoms, as well as to prevent the development of osteoporosis by reducing bone loss in the immediate postmenopausal period (Bush, 1992). Observational studies indicate that estrogen, alone or in combination with progestin, is associated with a 50 percent reduction in the risk of cardiovascular disease. Oral (as opposed to transdermal) estrogens alter plasma lipoproteins, raising levels of high-density lipoprotein cholesterol (''good" cholesterol) and lowering levels of low-density lipoprotein cholesterol (Matthews et al., 1989; Stampfer et al., 1991; NIH Consensus Development Panel on Triglyceride, High-Density Lipoprotein, and Coronary Heart Disease, 1993). Combination hormonal therapy (estrogen and progestin together) may offer an advantage over estrogen alone in altering cardiovascular risk factors (Nabulsi et al., 1993). Studies of the effect of postmenopausal hormone use on the risk of stroke have produced conflicting results, some showing no effect, others showing a benefit, and still others showing an increased risk (Stampfer et al., 1991; Finucane et al., 1993). There are other risks associated with estrogen or combined hormonal therapies. An increased risk of endometrial cancer has been associated with estrogen use, although this risk diminishes with the addition of a progestin (Bush, 1992). Whether or not hormone replacement increased the risk of breast cancer is unresolved (Horton, 1992). Recent data suggest that the

OCR for page 84
--> tained from a sample to a broader population. These arguments are reflected, implicitly or explicitly, in other chapters, for example in the concerns that clinicians and patients may be confused about whether a treatment that has been tested only on men is also effective—and safe—for women with the same disease. These concerns about generalizability have merit, but the external validity of clinical trials, as a methodological issue, is an oxymoron. External validity per se derives from drawing a representative sample of the population of interest. Other types of clinical studies that use survey techniques to sample randomly (say, from all Medicare patients hospitalized for gall bladder surgery) might claim to have external validity for describing the entire population (i.e., all Medicare patients with hospitalization for gall bladder surgery, and perhaps non-Medicare patients as well). Clinical trials, however, do not take a random sample from a representative population. Instead, they screen volunteers to see if they meet the selection criteria (see below) and then randomize the participants into treatment or nontreatment subgroups. This kind of design focuses instead on achieving internal validity—that is, how consistently and how well the treatment works. In this sense clinical trials, as a design, cannot truly speak to external validity issues; nevertheless, they can speak to gender differences in treatment effect, and in that sense can contribute to the knowledge and understanding of whether women and men differ in their responses to treatment. Homogeneity Versus Heterogeneity In order to assess the effect of treatments, clinical investigators try to reduce or eliminate sources of variance that are under their control, although they cannot achieve the same degree of control as the laboratory scientist. Hence, they construct criteria for the selection of subjects, criteria that are intended to reduce variance by recruiting the most homogeneous sample possible. From the narrow perspective of a single trial, the more homogeneous the population, the better. From this perspective, heterogeneity holds no intrinsic value and is simply the end result of not having been able to achieve the desired degree of homogeneity because of the costs involved (measured by effort or by dollars) or because of other limitations or constraints. The smaller the anticipated sample size of the trial, the more important it is to recruit a homogeneous sample with regard to factors known to affect treatment. There are two reasons for this:  Variations in the population enrolled in a small trial can have greater effect on the results than in a large trial. Randomization is more likely to do its "job" in a large trial than in a small one: a "bad break" in the randomiza-

OCR for page 84
--> tion process in relation to patient characteristics is more likely to be "confounding" in a small trial than in a large one. The ability to adjust for differences in the composition of the study groups (e.g., through regression techniques) is limited if not completely precluded in a small trial, whereas it is the method of choice for adjustment in the large trial. Exclusions from Trials As a general rule, the list of exclusions should be as short as possible, and each exclusion should be justified on medical or scientific grounds. The list should be pared by an active process of review and challenge prior to the start of the trial, and the list should be periodically reviewed during the trial for possible further trimming. Typical reasons for excluding patients from clinical trials include any or all of the following: Disease stage (the disease is too advanced or not advanced enough for the treatment being studied; prognosis inconsistent with treatment). Clinical contraindications (e.g., allergy to one of the study drugs). Regulatory or ethical restrictions (study drug not approved for use in children; subject is incapable of giving valid informed consent to participation, for example). Compliance considerations (e.g., history of illegal intravenous drug use in a drug trial requiring an indwelling line for drug infusion). Variance control (including age restrictions because of anticipated differences in response to treatment by age group). In every instance, exclusions should be based on strong supporting data in published documents. The first four reasons, assuming that they are based on well-reasoned evidence, can all be defended as being important for the validity of the study. The last general reason for exclusion, variance control, is where gender restrictions have been most often imposed (to achieve homogeneity) without actual evidence to suspect gender differences. Within broad limits, selection criteria related to the disease under scrutiny are sufficient to ensure scientifically based homogeneity; the addition of demographic variables adds little, particularly in treatment trials. In general, there are two justifications for excluding subjects on the basis of demographic characteristics such as gender, race, ethnicity, or age: (1) when these demographic statuses serve as surrogates for some other risk factor (as would be needed in a risk-concentration design) or (2) when the disease or condition being treated is concentrated in a particular subgroup (e.g., sickle cell anemia in African Americans or breast cancer in women). The mere fact that the number of any demographic subgroup enrolled in a

OCR for page 84
--> study would be too small to permit statistically valid subgroup analyses (such as women in a study of MIs) does not constitute a valid scientific reason for their exclusion. The arguments that the ability to detect the main effect because of the added variance arising from inclusion of both genders are generally without scientific validity; in most cases, the study group can be "homogenized" by carrying out statistical analyses taking gender and other factors into account. Age-based exclusions These issues can be illustrated by examining the age- and gender-based rationales for exclusions from several commonly cited male-only trials of conditions and treatments applicable to both men and women. In the Coronary Drug Project, for example, the rationale for the lower age limit of 35 had to do with the epidemiology of heart disease (that is, women and men alike experience relatively few MIs before that age) and the medical postulate that MIs occurring before 35 may be different from those occurring later in life (CDPRG, 1973). In the Multiple Risk Factor Intervention Trial (MRFIT), the rationale for the lower age limit of 35 was based on compliance-related considerations (MRFIT Group, 1977). Because the treatments involved sustained lifestyle changes, the study excluded the portion of the population (under 35) believed to be problematic in making and sustaining dietary changes in cholesterol intake. The rationale for upper age limits in adult populations is different. Often they are imposed in prevention trials to ensure sufficient remaining lifetime to allow the treatment to have an effect or to concentrate on the portion of the population believed able to benefit from the treatment. Such exclusions, however, should not be imposed without reference to the duration considered necessary, on average, to accrue treatment benefits, and they should not be used without an understanding of conditional life expectancies. Generally, most people well enough to walk into a clinic and enroll in a secondary prevention trial have several years of remaining life, regardless of their age. In 1989, for example, the conditional life expectancy at age 70 was 12.1 years for a white male, 11.0 for a black male, 15.3 for a white female, and 13.9 for a black female; the corresponding figures at age 85 were 5.3 for a white male, 6.5 for a white female, 5.6 for a black male, and 6.7 for a black female (U.S. Bureau of the Census, 1992). Gender-based exclusions More germane to this report are the rationales underlying the exclusion of women from two large preventive clinical trials: the Physicians' Health Study and MRFIT. The primary reason for excluding women from the Physicians' Health Study was the gender mix of the physician cohort approached for study (approximately 90 percent male); the number of women in the cohort was not large enough for a gender-bytreatment interaction analysis. The investigators' reservations regarding the

OCR for page 84
--> ability to perform such an analysis were sound, but the enrollment of a specified subgroup does not obligate them to perform interaction analysis for that subgroup, nor does the analysis need to be definitive if performed. If nothing else, the sign and size of any difference observed for women would have provided some general indication of whether the result obtained in men is consistent with that observed in, and thus generalizable to, women. Another possible argument for excluding women from clinical trials is that of efficiency: is the increase in information gained proportionate to the increased costs of including women? If including women would have resulted in a 10 percent increase in person-years of follow-up information, for example, then to justify excluding them it would have been necessary to show that adding women would have increased costs by more than 10 percent. In the case of the Physicians' Health Study, including women would probably have added valuable follow-up information without adding disproportionate costs. Most of its costs had to do with its coordinating center and other central functions; the cost for screening, treatment, and follow-up of each participant were relatively low. The basis for the gender exclusion in MRFIT is more compelling on both scientific and practical grounds. Screening costs were higher than those of the Physicians' Health Study because it was necessary to find people with a defined risk profile based on smoking behavior, cholesterol level, and blood pressure. The assessment of eligibility required the collection and analysis of blood, measurement of blood pressure, and an interview to characterize smoking behavior. Nearly 362,000 men were screened to find the 12,866 men enrolled into the trial; including women might have produced valuable information, but at a substantial additional cost. Subgroup Analysis In the context of any clinical study, an interaction is a relationship in which the response to treatment is moderated or influenced by some other variable or variables. The variable may be a demographic characteristic (e.g., gender) or some other baseline characteristic or variable (e.g., nonsmoker). An interaction effect is said to exist when the treatment effect differs depending on which status of a demographic or baseline variable a person exhibits—such as being a woman or man. A qualitative interaction is one in which the direction or sign of the relationship depends on the value assumed by the demographic or baseline variable of interest (e.g., one in which the treatment effect is beneficial for males but is harmful for females). A quantitative interaction is one in which the sign or direction of the relationship is the same, but the magnitude of the effect is different. A subgroup analysis in this context is any comparison of treatments within a given population, using one or more demographic or baseline characteristics

OCR for page 84
--> to define subgroups. The characteristics of interest here are those having to do with biological differences associated with gender, race, ethnicity, and age (IOM, 1993a). Most of the researchers who design and analyze clinical trials are skeptical of any observed differences in treatment effects among demographic subgroups unless there is a clear scientific rationale (grounded in biological differences) or until the differential outcome has been replicated in other studies. The underlying assumption is that treatments that work well in one demographic subgroup are likely to work well in another, unless there is a biologically plausible reason for believing otherwise. This argues for the inclusion of women, but not necessarily for the proactive recruitment of sufficient numbers to be able to perform meaningful subgroup analyses. Nevertheless, in the interests of adding to the knowledge of whether there are any unsuspected differences by gender (or other key demographic characteristics), the committee believes that examination of subgroups—where feasible given the number of people enrolled in each subgroup—is to be encouraged. Alternatives to Clinical Trials Strategies other than clinical trials are available to help devise hypotheses about the differential response of men and women to medical interventions. These strategies may be significantly less costly than large-scale clinical trials that include sufficient numbers of men and women to detect gender differences in response. Meta-analysis is one such strategy that is inherently inexpensive and particularly useful for detecting subtle associations between interventions and outcomes and between demographic characteristics and drug effects (IOM, 1993a). Outcomes research, which involves systematic study of the health impact of an intervention, is another such strategy. Pharmacoepidemiologic research is a specific type of outcomes research designed specifically for the study of drug effects in user populations. Meta-analysis refers to a set of quantitative techniques for combining data from different studies of the same or similar phenomena. From information obtained from each study, a synthesis is made that may produce a much stronger conclusion than any of the studies by themselves can provide. Meta-analysis can be used by clinical investigators to detect significant differences between treatment and control groups where sample sizes in individual studies were too small to allow the detection of statistically significant effects. At the same time, meta-analysis may reveal through averaging that an effect that appeared to be significant in one study is actually less significant. Meta-analysis also allows investigators to detect contradictions or discrepancies among groups of studies. Faced with a collection of studies in a particular area of research, investigators may analyze

OCR for page 84
--> and compare subgroups of studies with, for example, divergent findings to detect mediating factors of study design, treatment, context, measurement, or analysis that otherwise may not have appeared noteworthy (Pillemer and Light, 1980). The utility of meta-analysis techniques depends on assumptions made about similarities among grouped studies. The pharmaceutical industry has recently placed renewed emphasis on the use of nonexperimental, observational, epidemiologic techniques—or pharmacoepidemiologic techniques—as alternatives (and sometimes complements) to randomized controlled clinical trials. Postmarketing surveillance is one of several pharmacoepidemiologic techniques employed to study the effects of drugs in uncontrolled, "real life" settings, and in larger numbers of people than can be included in the drug development process. Pharmacoepidemiologic studies avoid some of the important shortcomings of clinical trials, namely the introduction of intervention and observation effects, and the exclusion of effects resulting from the usage of concomitant medications, presence of other illnesses, and lack of patient compliance (Tilson, 1993). The coincident revolutions in computer technology and health care management have yielded an important new resource for pharmacoepidemiology: large, automated, multipurpose databases of patient information collected by health maintenance organizations. In some cases (e.g., Group Health Cooperative of Puget Sound in Seattle, Washington), these databases contain medical records for as many 300,000 persons in a particular geographic area. Such large databases permit the epidemiologist to collect information on all drug exposures (because the databases include bills or records of disbursement of drugs from pharmacies) and all major medical outcomes (because the databases include bills or statements of diagnoses from clinics), and to take into account major stratifying variables (Tilson, 1993). Where calculated rates of adverse events among users tend to unreliable, largely as a result of underreporting and differential reporting, a structured epidemiologic study can be a powerful tool for quantitatively studying unexpected adverse events. Although pharmacoepidemiologic studies are generally less costly than clinical trials, these studies can also be complex and expensive, particularly if the risk of an adverse event is small (and the population needed to detect events therefore large); if the event itself is subtle (and monitoring is complex); and if the period of potential risk is long (and the population must be studied for a long time) (Tilson, 1993). However, in cases in which it is impossible or unacceptable to conduct randomized controlled clinical trials (e.g., in persons dying of HIV disease), these studies provide a critical alternative. Recently, the pharmaceutical industry has experimented with new techniques designed to expedite and enhance the results of drug development.

OCR for page 84
--> One of these techniques, which may be useful in detecting differential responses to drugs between men and women (as well as other subgroup differences), is the pharmacokinetic screen. The pharmacokinetics of a drug refer to the drug's absorption, distribution, and metabolism in the body, as well as the drug's excretion from the body. A pharmacokinetic screen is a technique that can be used during drug development to infer the influence of demographic factors such as age and gender on pharmacokinetics and to suggest the likelihood that a drug-drug or drug-disease interaction will (or will not) occur. The screening process involves the analysis of drug levels in members of specific subgroups at designated points (e.g., before and after dosage) throughout the course of a Phase III trial. Conducting a pharmacokinetic screen during drug development adds little extra cost to the development process, because the necessary data are collected from patients already participating in trials. CONCLUSIONS AND RECOMMENDATIONS In general, the committee's findings are compatible with the NIH mandate for broader inclusion of women and ethnic and racial groups in clinical studies, but with some important caveats, summarized here. The difficult policy issues about including women in clinical studies do not arise when a disease or condition is exclusive to men. Clearly, they do not arise when the study is based on administrative records such as insurance claims, because it is relatively easy to sample patients of both genders. Instead, they attend studies that require large resources to administer and collect new data. For these reasons we focus our concluding remarks on clinical trials of treatment effects for diseases that affect both women and men—namely treatment trials. The committee finds that the weight of scientific evidence, as well as practical considerations, supports the inclusion of both genders—and indeed all kinds of demographic subgroups—wherever possible. The most compelling scientific reasons for exclusion are found in investigations of diseases, conditions, or risk factors (including behavior) that are highly concentrated in a single gender. Ethical concerns such as an incapacity to give valid informed consent can provide compelling reasons for exclusion irrespective of the presence of scientific reasons, but these are discussed elsewhere in this report. This is not to say that there are no significant gender-specific diseases or treatment effects, nor do we mean to argue that sufficient attention has been paid to the possibility of gender-specific differences. We support the need to examine these issues systematically where they are based on well-grounded scientific hypotheses, and we support attempts to encourage scientists and clinicians to consider and pursue such gender-related hypoth-

OCR for page 84
--> eses. We also acknowledge, however, that most treatments and most diseases do not differ significantly in their effect by gender. This observation reinforces rather than reduces the justification for a principle of inclusion: if indeed most treatment effects in the setting of treatment trials do not differ by gender, then it is reasonable for treatment trials to include both genders. Scientific considerations suggest that the overarching principle should be inclusion. Some would argue that excluding women is justified in a trial where there is no anticipated difference in how women and men respond to a treatment but where the disease is less common among women. These arguments rest on a false assumption that women's presence diminishes homogeneity and thereby lessens the ability to observe the main effect of the treatment (i.e., whether the treatment is effective for any subject). Person-years of follow-up are person-years of follow-up whether they are female or male years, unless the researchers have plausible hypotheses about gender differences in response. And if they do have convincing hypotheses about qualitative gender-specific differences, then this too argues for including both genders, but in sufficient numbers to test for gender-specific results. When there are no anticipated treatment effects by gender, however, a policy that requires scientists to include sufficient representation of both genders to permit subgroup analyses would require, at a minimum, that clinical trials significantly increase their size (to detect the main effect in each group) and proportionately increase their expenses. The committee has two concerns in this case, one about the meaning of any information gained by ''data dredging," and the other about the appropriateness of spending finite research dollars to satisfy requirements that have little possibility of producing useful information. The NIH Revitalization Act of 1993 (P.L. 103-43) appears to be based on an intent that is consistent with the principles stated in this chapter, with one important exception. Before excusing a clinical trial from implementing a design with sufficient power to detect subgroup interactions, the act requires investigators to produce "substantial scientific data" of no gender (or race or ethnic) differences. Although the committee agrees with the intent, we believe that this criterion is too extreme given the nature of scientific evidence. Instead, we would suggest a requirement that applicants for funding provide a continuing review of the evidence pertaining to gender-specific effects, along with an assessment of the sample sizes (and costs) that would be necessary to support subgroup analyses. The committee is concerned that the policy has gone too far by insisting that each and every clinical trial be designed to ensure sufficient numbers of subjects of both genders to permit subgroup analyses. In an era of concern about the nation's resources, and about expenditures on health in particular, we argue that a trial-by-trial application of this requirement is nei-

OCR for page 84
--> ther good policy nor good science. Clinical trials should include both genders, but requiring scientists to enroll sufficient numbers to ensure the statistical power to detect unsuspected and implausible gender differences would produce little additional information at greatly increased cost. Instead, mechanisms are needed at the national level to ensure that more attention is paid to questions of justice and gender in the setting of research agendas, and at the study level to encourage the appropriate consideration of gender-related effects in clinical studies. The specific actions we suggest to accomplish these goals are discussed in Chapter 8; our general recommendations are as follows: The committee recommends that NIH commission a study to identify known gender differences in drug response. The committee recommends that investigators be attentive to factors associated with possible gender differences in drug response and design their studies accordingly. Further, NIH should commission a study that will assist investigators in their effort to detect such differences. The committee recommends that investigators avoid exclusions based on demographic characteristics. The committee recommends that investigators proposing research involving human subjects provide a reasonable review of the evidence and plausibility of gender-specific effects relevant to their research, and that studies be required to be designed with sufficient power to detect subgroup differences only when such a review indicates that such a design is warranted. When there is no information concerning possible gender differences, however, the investigator should, when feasible, include both genders in sufficient number to detect differences. Strategies other than clinical trials, (e.g., surveillance techniques) are available to help devise hypotheses about the differential response of men and women to medical interventions. These strategies may be significantly less costly than large-scale clinical trials that include sufficient numbers of men and women to detect gender differences in response. The committee recommends that NIH assist investigators in this effort by: (1) identifying, developing, and disseminating alternative methods for detecting or formulating hypotheses about gender differences and (2) providing guidance for the use of these methods by investigators and initial review groups.

OCR for page 84
--> REFERENCES Bush, T.L. 1992. Feminine forever revisited: Menopausal hormone therapy in the 1990s. Journal of Women's Health 1(1):1-4. CDPRG (Coronary Drug Project Research Group). 1973. The Coronary Drug Project: Design, Methods and Baseline Results. American Heart Monograph No. 38, New York: American Heart Association. Everitt, D., and Avorn, J. 1986. Drug prescribing for the elderly. Archives of Internal Medicine 1986:146(12):2393-2396. Finucane, F.F., Madans, J.H., Bush, T.L., Wolf, P.H., and Kleian, J.C. 1993. Decreased risk of stroke among postmenopausal hormone users: Results from a national cohort. Archives of Internal Medicine 153:73-79. GAO (General Accounting Office). 1992. Women's Health: FDA Needs to Ensure More Study of Gender Differences in Prescription Drug Testing. GAO/HRD-93-17. Washington, D.C.: GAO. Gillette, J.R. 1977. Factors that affect drug concentration in maternal plasma. Pp. 35-77 in Handbook of Teratology, Vol. 3, J.G. Wilson and F.C. Fraser, eds. New York: Plenum. Horton, J.A., ed. 1992. The Women's Health Data Book. Washington, D.C.: The Jacobs Institute of Women's Health. Hytten, F.E., and Chamberlain, G. 1980. Clinical Physiology in Obstetrics. Oxford, England: Blackwell. IOM (Institute of Medicine). 1993a. Inclusion of Women in Clinical Trials: Policies for Population Subgroups. Washington, D.C.: National Academy Press. IOM (Institute of Medicine). 1993b. Women and Drug Development. Report of a workshop, Forum on Drug Development. Washington, D.C.: National Academy Press. Jensvold, M.F., Reed, K., Jarrett, D.B., and Hamilton, J.A. 1992. Menstrual-cycle related depressive symptoms treated with variable antidepressant dosage. Journal of Women's Health 1(2):109-115. Matthews, K.A., Meilahn, E., Kuller, L.H., Kelsey, S.F., Caggiula, A.W., and Wing, R.R. 1989. Menopause and risk factors for coronary heart disease. New England Journal of Medicine 321(10):641-646. Mattison, D. 1986. Physiological variations in pharmacokinetics during pregnancy. Pp. 37-102 in Drug and Chemical Action in Pregnancy, S. Fabro and A.R. Scialli, eds. New York: Marcel Dekker, Inc. Mattison, D.R. 1990. Transdermal drug absorption during pregnancy. Clinical Obstetrics and Gynecology 33:718-727. Mattison, D.R., and Jelovsek, F.R. 1991. Environmental and occupational exposures. Pp. 91109 in Reproductive Risks and Prenatal Diagnosis, M.I. Evans, ed. Norwalk, Conn.: Appleton & Lange. Mattison, D.R., Malek, A., and Cistola, C. 1992. Physiological Adaptations to Pregnancy: Impact on Pharmacokinetics. Pp. 81-96 in Pediatric Pharmacology: Therapeutic Principles in Practice, 2nd ed., J. Aranda and S.J. Yaffe, eds. Philadelphia: W.B. Saunders. McGrath, E., Keita, G.P., Strickland, B.R., and Russo, N.F. (eds). 1990. Women and Depression: Risk Factors and Treatment Issues, Final Report of the American Psychological Association's National Task Force on Women and Depression. Washington, D.C.: American Psychological Association. Mosher, W.D. 1990. Contraceptive practice in the United States, 1982-1988. Family Planning Perspectives 22:198-205. Multiple Risk Factor Trial Group. 1977. Statistical design considerations in the NHLI Multiple Risk Factor Intervention Trial (MRFIT). Journal of Chronic Disease 30:261-275. Nabulsi, A.A., Folsom, A.R., White, A., Patsch, W., Heiss, G., Wu, K.K., and Szklo, M. 1993.

OCR for page 84
--> Association of hormone replacement therapy with various cardiovascular risk factors in postmenopausal women. New England Journal of Medicine 328(15):1069-1075. National Center for Health Statistics. 1985. Current Estimates from National Health Interview Survey. Vital and Health Statistics. Atlanta, Ga.: National Center for Health Statistics, Centers for Disease Control, Series 10, No. 160. National Center for Health Statistics. 1993. Chartbook on Health Data on Older Americans, 1992. Atlanta, Ga.: National Center for Health Statistics, Centers for Disease Control, Series 3, No. 29. National Medical Expenditure Survey. 1987. Prescribed Medicines: A Summary of Use and Expenditures by Medicare Beneficiary. Dept. H, 45, Research Findings 3. National Women's Health Resource Center. 1990. Forging a Women's Health Research Agenda. Washington, D.C.: National Women's Health Resource Center. NIA (National Institute on Aging). 1992. Bound for Good Health: A Collection of Age Pages. Bethesda, Md.: NIA. NIH (National Institute of Health). 1993. Consensus Development Panel on Triglyceride, High-Density Lipoprotein, and Coronary Heart Disease. Triglyceride, high-density lipoprotein, and coronary heart disease. Journal of the American Medical Association 269(4):505-510. NIH (National Institutes of Health). 1992. Opportunities for Research on Women's Health. Proceedings of a conference in Hunt Valley, Md., Sept. 4-6, 1991. Bethesda, Md.: NIH. NIH (National Institutes of Health). Advisory Committee on Women's Health Issues. 1988. Women's Health Issues Research Report, FY 1985-1987. Bethesda, Md. NRC (National Research Council). 1993. Pesticides in the Diets of Infants and Children. Washington, D.C.: National Academy Press. Olshansky, S.J., Carnes, B.A., and Cassel, C.K. 1993. The aging of the human species. Scientific American (April):46-52. Pelkonen, O. 1980. Biotransformation of xenobiotics by the fetus. Pharmacology and Therapeutics 10:26. Pillemer, D.B., and Light, R.J. 1980. Synthesizing outcomes: How to use research evidence from many studies. Harvard Education Review 50:176-195. Rodin, J., and Ickovics, J.R. 1990. Women's health: Review and research agenda as we approach the 21st century. American Psychologist (September): 1018-1034. Silvaggio, T., and Mattison, D.R. 1993. Comparative approach to toxicokinetics, In: Occupational and Environmental Reproductive Hazards: A Guide for Clinicians, M. Paul, ed. Baltimore, Md.: Williams & Wilkins. Stampfer, M.J., Coditz, G.A., Willett, W.C., Manson, J.E., Rosner, B., Spaizer, F.E., and Hennekens, C.H. 1991. Postmenopausal estrogen therapy and cardiovascular disease: Ten year follow-up from the nurse's health study. New England Journal of Medicine 325(11):756-762. Tanner, L.A., Baum, C., Prela, C.-M., and Kennedy, D.L. 1989. Spontaneous adverse reporting in the elderly for 1985. Journal of Geriatric Drug Therapy 3(3):31-54. Teichmann, A.T. 1990. Influence of oral contraceptives on drug therapy. American Journal of Obstetrics and Gynecology 163:2208-2213. Tilson, H. 1993. Observational studies making an overdue and much needed comeback: Alternatives and complements to randomization. In: Clinical Trials and Statistics: Proceedings of a Symposium. Washington, D.C. : National Academy Press. U.S. Bureau of the Census. 1992. Statistical Abstract of the United States: 1992 (112th ed.). Washington, D.C. Weatherall. D.J., et al., eds. 1988. Oxford Textbook of Medicine, Volume 2. New York: Oxford University Press. Yonkers, K.A., Kando, J.C., Cole, J.O., and Blumenthal, S. 1992. Gender differences in pharmacokinetics and pharmacodynamics of psychotropic medication. American Journal of Psychiatry 149(5):587-595.